Why does tree conditions stopped working

[DenisSvirin]

Right now, I have custom_objective, where right part of a tree is divided into to parts: left (tree_att) and right (power).
I have constrains on power to be constant.

I need to calculate derivative of this tree, but I have to change it; to look like tree_att ^ (2 * power).
If I do this like in the code below, trees in the hall of fame stop following constrains, to be particular power is no longer a constant.
Without changing a tree, there are no problems.

Why can this problem occur?

    # Two body part
    E_two_body = tree

    E_two_body.degree != 2 && return L(Inf)

    # E repulsion
    tree_rep = E_two_body.l

    # E attraction, power
    tree_att_p = E_two_body.r

    tree_att_p.degree != 2 && return L(Inf)

    # E attraction
    tree_att = tree_att_p.l

    # power
    power = tree_att_p.r
    power.constant != true && return L(Inf)

    ...

    eval_power = power.val
    tree_att_p.op = 5 #options = [+, -, *, /, ^]
    tree_att_p.r.val = 2 .* abs(eval_power)

[MilesCranmer]

I think you need

-    power.constant != true && return L(Inf)
+    (power.degree != 0 || power.constant != true) && return L(Inf)

because the .constant field is some random value if it's an operator node. So for it to be a constant node it has to be .degree == 0 AND .constant == true.

Also one other tip is that rather than returning L(Inf) for all of the constraints, you should return lesser and lesser penalties so that the evolutionary algorithm gets a sense of direction. e.g., for the first L(Inf), make it L(100_000) (since that is the most important constraint to get right). The second L(Inf), make it L(50_000), since at least then it got the first constraint. Then L(25_000). And so on.

---

Also I would do

    eval_power = power.val::T

for improved speed

[DenisSvirin]

That works! Thanks